Modeled large-scale warming impacts on summer California coastal-cooling trends



[1] Regional Atmospheric Modeling System (RAMS) meso-meteorological model simulations with a horizontal grid resolution of 4 km on an inner grid over the South Coast Air Basin of California were used to investigate effects from long-term (i.e., past 35 years) large-scale warming impacts on coastal flows. Comparison of present- and past-climate simulations showed significant increases in summer daytime sea breeze activity by up to 1.5 m s−1 (in the onshore component) and a concurrent coastal cooling of average-daily peak temperatures of up to −1.6°C, both of which support observations that the latter is an indirect “reverse reaction” to the large-scale warming of inland areas.

1. Introduction

[2] California complex-topography coastal-areas are influenced by mesoscale coastal flows, initiated by near-surface (herein referred to as “surface”) temperature differences between land and ocean that produce onshore daytime sea breezes and offshore nighttime land breezes, both of which result from land-sea temperature differences [Pielke, 1984]. Related phenomena include return flows aloft, marine boundary layers (MBLs) (all acronyms are defined in Table 1), and associated inversions, low level jets (LLJs), upslope and downslope winds, and topographic channeling [Mahrer and Pielke, 1977]. These phenomena depend on time of year, latitude, sea surface temperatures (SSTs), planetary boundary layer (PBL) depth and stability, and on factors that alter land-surface energy balances (e.g., clouds, land use, albedo, and soil moisture). The South Coast Air Basin (SoCAB) sea breeze has been documented as consisting of a shallow cool coastal marine boundary layer (MBL) that deepens and warms [Edinger, 1959] as it advances inland until its frontal inversion weakens and finally erodes [Edinger, 1963].

Table 1. Acronyms and Abbreviations
2-Dtwo dimensional
3-Dthree dimensional
AVHRRAdvanced Very High Resolution Radiometer
BATSBiosphere-Atmosphere Transfer Scheme
BCsboundary conditions
C-CAPCoastal Change Analysis Program
CCNYCity College of New York
COOPcooperative network
CPUcentral processing unit
CSUColorado State University
DEMdigital elevation model
DTRdiurnal temperature range
ERSSTextended reconstructed SST
EROSEarth Resources Observation System
GCgeneral circulation
GCMglobal circulation model
GHGgreenhouse gases
ICOADSInternational Comprehensive Oceanographic and Atmospheric Data Set
LEAF-3Land Ecosystem Atmosphere Feedback Model version 3
LLJlow level jet
LSTlocal standard time
LTlocal time
LULCland use–land cover
MBLmarine boundary layer
METARMeteorological Terminal Aviation Routine
NCDCNational Climate Data Center
NCEPNational Centers for Environmental Prediction
NDVInormalized difference vegetation index
NOAANational Oceanic and Atmospheric Administration
OGEOlson Global Ecosystem
PBLplanetary boundary layer
PDOPacific Decadal Oscillation
RAMSRegional Atmospheric Simulation System model
SCUSanta Clara University
SFBASan Francisco Bay Area
SoCABSouth Coast Air Basin
SSTsea surface temperature
TKEturbulent kinetic energy
USGSUnited States Geological Survey
UTCuniversal time coordinated

[3] These SoCAB mesoscale phenomena have also been modeled; e.g., Ulrickson and Mass [1990] found that summer daytime upslope flows ventilate the SoCAB by merging with and strengthening sea breeze flows. These upslope flows converge at inland mountain tops, producing strong vertical motions, which create upper-level return offshore-directed flows. Polluted air is prevented from exiting the SoCAB (in the pass between the San Gabriel and San Bernardino Mountains) by opposing upper level easterly flows associated with mesoscale highs north of that gap [Boucouvala et al., 2003].

[4] Observational analyses have indicated that cooling of summer (i.e., June to August, JJA) maximum temperature (Tmax) values in coastal California over the last several decades could be due to increased: irrigation [Bonfils and Lobell, 2007], coastal upwelling [Bakun, 1990], stratus cover [Nemani et al., 2001], and/or urban cool-islands [LaDochy et al., 2007]. Lebassi et al. [2009], however, showed that this cooling from 1970 to 2005 was due to a “reverse reaction” to the large-scale warming of inland areas, which increases a sea breeze activity that overwhelms the warming in low-elevation SoCAB coastal areas. Irrigation is not a factor in these coastal areas, but increased upwelling, cold air advection, and stratus increases probably contribute to this observed cooling.

[5] The JJA results of Lebassi et al. [2009] for all 253 California Cooperative Observed Program (COOP) sites together showed increased (0.15°C decade−1) average (Tave) values; asymmetric warming, as minimum temperature (Tmin) values increased faster than Tmax-values (0.27 versus 0.04°C decade−1); and thus decreased diurnal temperature range (DTR) values (−0.23°C decade−1). While JJA nighttime SoCAB minimum temperatures Tmin showed expected large-scale warming, their Tmax-values exhibited a complex spatial pattern, with a domain-averaged cooling (up to −0.99°C decade−1, although only the −0.6°C decade−1 isopleth can be shown conveniently) in low-elevation coastal-areas open to marine air penetration and warming (up to 0.4°C decade−1) at inland areas, which suggests that the inland warming resulted in increased coastal sea-breeze activity. Note that the spatial distribution of the SoCAB Tmax-trends (Figure 1) shows a maximum cooling not along the coast, but somewhat inland in the San Fernando Valley, implying that sea breezes were not as frequent 35 years ago in the Valley. Lebassi et al. [2010] showed that observed per capita energy consumption for summer cooling in the coastal-cooling areas decreased during the period, while it increased in inland warming areas.

Figure 1.

Spatial distribution of observed SoCAB 2-m summer maximum-temperature trends (°C decade−1) for 1970–2005; arrows indicate predominant summer daytime surface-flow patterns; blue, red, and black indicate cooling, warming, and no-change isopleths and station locations, respectively; dashed isopleths are extrapolated; and statistical significance values of >99%, between 95 and 99%, between 90 and 95%, and <90% are represented, respectively, by full-colored, half-colored, plus sign in circles, and open circles (from Lebassi et al. [2009], © 2009 American Meteorological Society (AMS).)

[6] General Circulation Model (GCM) simulations by Cayan et al. [2008] project a total 21st century warming of 2–5°C for California surface temperatures. Downscaled regional climate modeling by Kueppers et al. [2007] showed western-U.S. irrigation lowering average temperature (Tave) and Tmax values at rates comparable to the increases from large-scale warming. Similar modeling by Miller et al. [2008] showed California Tave warming rates of 2.1–4.5°C, while ensemble downscaling by Maurer [2007] of 1950–99 JJA median temperatures showed warming rates that decreases from 0.13°C decade−1 in inland California to 0.08°C decade−1 at coastal areas north of the SoCAB. While this showed coastal influences, its coarse spatial resolution could not sufficiently resolve local topographic features that affect mesoscale circulations, and thus local temperature trends. Local-scale modeling by Lobell et al. [2006] showed California temperature changes generally determined by large-scale warming, but with large land-use changes locally dominant.

[7] Similar modeling by Snyder et al. [2003] and Duffy et al. [2007] showed increased large-scale warming strengthens onshore pressure and temperature gradients, as land areas warmed faster than ocean areas due to thermal differences. Thus enhanced alongshore winds strengthened upwelling, which further increased onshore temperature gradients. McGregor et al. [2007] observed this effect over coastal northwest Africa, while Bakun [1990] hypothesized a similar scenario to explain observed 30 year increases of California upwelling.

[8] Comarazamy et al. [2010] analyzed surface-atmospheric interactions in the tropical coastal city of San Juan, Puerto Rico by the use of the Regional Atmospheric Modeling system (RAMS) for two periods, i.e., past (1950–1955) and present (2000–2005), to study the individual and combined impacts of urbanization and large-scale climate change. Results showed that in San Juan, both effects contributed to warming, with urbanization having a larger impact.

[9] While previous studies generally attributed observed JJA asymmetric warming in coastal California during the last three decades to increases in cloud cover, SSTs, upwelling, land use changes (i.e., urbanization, and/or irrigation). The current study uses the RAMS mesoscale meteorological model to quantify large-scale warming effects on SoCAB summer sea breeze patterns and thus on Tmax trends.

2. Methodology

2.1. Model Description

[10] The overall goal of the current high-resolution RAMS mesoscale meteorological numerical-model simulations of current (i.e., 2001–05) and past (i.e., 1966–70) SoCAB JJA sea breeze wind and temperature patterns is to relate their changing patterns to the observed concurrent inland-warming and coastal-cooling patterns of Lebassi et al. [2009]. The simulations incorporate the combined effects from large-scale warming and SST-changes.

[11] An outer grid with a horizontal grid spacing of 16 km, and an inner grid of 4 km, are specified (Figure 2) to capture larger scale (i.e., General Circulation (GC) and synoptic) forcings and to further resolve more complex smaller mesoscale phenomena, respectively. Simulations of 35 years of summer months are not feasible on such high resolution grids, because of the required large computational (i.e., one hour Central Processing Unit (CPU) time for one-day simulation-time with a 24 processor cluster) and storage resources. The current simulations are thus only for five consecutive summers during both the current and a past period. Resulting temperature and wind difference-fields were then calculated by subtraction of present minus past values.

Figure 2.

Outlines of Grid-1 and -2, where colors indicates topographic heights (m).

[12] RAMS (Version 6.0), developed at Colorado State University (CSU), solves the Reynolds-averaged quasi-Boussinesq, non-hydrostatic, primitive equations on a Polar stereographic map projection [Tripoli and Cotton, 1982]. It uses terrain-influenced sigma coordinates and an Arakawa-C staggered grid, on which thermodynamic and moisture variables are defined at grid-volume centers, with velocity components on grid-face centers perpendicular to each flow component [Mesinger and Arakawa, 1976].

[13] Two fixed two-way interactive grids are centered over the SoCAB, with 30 vertical layers, stretched from a spacing of 0.03 to 1.2 km in the first 7.5 km (to increase near-surface resolution). The model uses time-split differencing [Pielke, 1984], as well as variable-field initialization- and update-processes, in which gridded 3-D fields of wind, potential temperature, and relative humidity values from a large-scale (i.e., National Centers for Environmental Prediction (NCEP) operational-analysis, in this application) model are assimilated at 6 h intervals via Newtonian relaxation [Davies, 1983].

[14] The cloud microphysics formulation of Walko et al. [1995] and Meyers et al. [1997] was used; as it includes complex parameterization of many ice species. A cumulus parameterization was not used, as both grid resolutions were at or below the 9-km resolution needing such formulations. Vertical diffusion coefficients were computed from the 2.5-level closure of Mellor and Yamada [1982], which employs a prognostic turbulent kinetic energy (TKE) equation. Short- and long-wave radiation were calculated following Mahrer and Pielke [1977], in which simulated water vapor and specified constant CO2 concentrations affect radiative flux divergence optical-path calculations via their effects on atmospheric-emissivity.

[15] The CO2 concentration was assumed uniform in both domains at 330 ppm from the surface to 40 km, while downwelling long-wave flux at the model top was set to zero. This value is comparable with the summer 329 ppm value observed by Keeling and Whorf [2004] at La Jolla, California in the early 1970s, while it underestimates the 2001 value of 369 ppm. Radiative effects from long-term global aerosol- and CO2-increases, accounted for via the NCEP boundary conditions (BCs), are ingested into RAMS every 6 h. Inner-domain radiative effects from aerosol and CO2 trends are not accounted for, but should be small over the five-summer simulation periods.

[16] The 1966–1970 and 2001–2005 periods were selected for simulation, as they showed similar large-scale Pacific Decadal Oscillation (PDO) climate-variability factors. Strong winter El Niños produce decreased coastal upwelling in southern California, while strong winter La Niñas produce the reverse [LaDochy et al., 2007]. The more important multidecadal index influencing California summer climate, however, is the spring (March, April, and May, MAM) PDO, evaluated from north Pacific SSTs. The JJA past- and present-periods were thus selected based on similarities between the trends of the anomalies of this index and those of coastal California JJA Tave-values (Figure 3), i.e., both periods generally show small upward trends in both variables (

Figure 3.

Five-year running average March, April, and May coastal-California Pacific Decadal Oscillation temperature-anomaly (gray dashed line) and average 2-m temperature anomaly (black line), both in °C.

[17] The following JJA simulations were thus carried out over five consecutive JJA-periods, as this was deemed sufficient to evaluate statistical differences: Run 1 (“Present”): current (2001–2005) climate and current (2002) urbanization; Run 2 (“Past”): past (1966–1970) climate and current (2002) urbanization.

[18] Run-1 results thus provides insight into summer-averaged current coastal SoCAB temperature and sea breeze flow patterns, while Run-2 provides insight into these processes under past-climate conditions. Run-1 minus Run-2 results thus provides insight into the effects of global warming on coastal temperatures and sea breeze flow patterns.

2.2. Initialization, BCs, and Statistical Evaluation

[19] RAMS was initialized with NCEP gridded data sets, interpolated (offline) to its grid by its internal isentropic analysis package. NCEP data contains the following 4-D fields on horizontal pressure surfaces at horizontal increments of 2.5 deg: horizontal velocity components, temperature, geopotential height, and relative humidity. These fields generate 3-D assimilation fields every 6 h during the execution [started at 2400 universal time coordination (UTC), i.e., 1700 local time (LT) in summer] that nudges coarse-grid lateral boundary-regions. Lateral BCs on the outer grid follow Klemp and Lilly [1978], a variant of Orlanski [1976], in which gravity-wave propagation speed computed for each model-cell are averaged vertically, with the single average-value applied over the entire vertical column. In this scheme, horizontal diffusion coefficients are computed as the product of horizontal deformation rate and length-scale squared [Smagorinsky, 1963].

[20] Initialization of RAMS also requires characterization of its surface BCs, via the following four input data sets, constructed “off line,” and then interpolated to its 2-D surface grid points:

[21] For topography, the U.S. Geological Survey (USGS) topographic heights at a resolution of 30 arc-sec (about 1 km) were obtained from the RAMS web site.

[22] Monthly JJA extended reconstructed SSTs (ERSSTs) were originally produced by statistical methods that allow for stable reconstructions from sparse data [Smith and Reynolds, 2003] by use of the international comprehensive oceanographic and atmospheric data set (ICOADS) data. Values are down-loaded from the national climate data center (NCDC) at a resolution of 1-deg (about 111 km) for the current-condition run and at 2-deg for the past-condition run.

[23] For the normalized difference vegetation index (NDVI), monthly values are obtained from the RAMS web site on a 30 arc-sec grid.

[24] For land use–land cover (LULC), advanced very high resolution radiometer (AVHRR) data are used in the 1-km resolution Olson [1994] Global OGE 94-class LULC classification scheme, available from the USGS earth resource observation system (EROS) Data Center [Lee, 1992]. Olson global ecosystem (OGE) LULC data are input into the RAMS biosphere-atmosphere transfer scheme (BATS), which condenses its 94 classes into 21. OGE data are used in the outer-domain for the current-period simulations, but the inner-domain uses the 30 m resolution, 38-class, 2000 data from the National Oceanic Atmospheric Administration (NOAA) coastal change analysis program (C-CAP) ( OGE LULC data are normally condensed to one urban and 20 rural classes in RAMS, but for the current simulations, two SoCAB urban LULC classes dominate: commercial and low-density residential; the one additional urban class was thus added to version-3 of the Land Eco-system Atmosphere Feedback (LEAF-3) model [Walko et al., 2005].

[25] LEAF-3 is used to calculate time-varying 2-D surface-temperature and -humidity BCs from linked prognostic surface-energy and -moisture balance equations for each of the following basic LULC “type” (i.e., group of similar classes): bare soil, plant-canopy covered soil, and urban; SST is assumed constant. Within any surface grid cell, 30-m patches are used to represent the heterogeneity of LULC classes. Once moisture and temperature values for each LULC category within each surface grid are calculated, heat and moisture fluxes are area-averaged within each cell, used within SBL-parameterizations as (constant with height) fluxes, and finally used as the lower BCs for the sub-grid diffusion schemes in the finite differenced prognostic PBL equations.

[26] As the inner domain is mainly urbanized, RAMS surface BCs were modified, to better account for urban processes, by the addition of a time- and spacing-varying anthropogenic heat flux (QA) for each urban sub-grid 30 m “patch.” In U.S. cities, QA is typically 60% from traffic sources and 40% from residential and industrial activities [Sailor and Lu, 2004]. The current effort implements a time-varying daily profile (Figure 4), which shows morning and afternoon peaks associated with rush hour traffic.

Figure 4.

SoCAB summer anthropogenic-heating diurnal- pattern (W m−2); data from Sailor and Lu [2004].

[27] RAMS look-up tables provide literature-values (as a function of LULC class) for the radiative, physical, vegetative, and thermal parameters associated with the surface-BC formulation. Values for several key parameters for urban and rural LULC classes (Table 2) show increased values of urban albedo and vegetation height.

Table 2. Selected RAMS LEAF-3 Lookup Table Input Values
LULC ClassSurface AlbedoVegetation Height (m)
Short grass0.260.3
Low intensity urban0.206.0
High intensity urban0.2020.0

[28] As is the spatial distribution of vegetation fraction is an important input-parameter, a technique was developed to determine it from high resolution Google™ earth-visible images as follows: (a) start with visible color Google map, (b) reduce map colors to 16, (c) count pixel values of each color in selected “typical” area for each urban class, (d) calculate fraction of green-color pixels to determine vegetation fraction, (e) reduce map colors to two (black and white), (f) count both the black and white pixels, (g) calculate fraction of white pixels to determine rooftop fraction, and (h) calculate street fractions as black fraction minus vegetation fraction. Only a vegetation-fraction value for each urban class can currently be input into the RAMS lookup table; further details can be found in work by Lebassi [2010].

[29] An evaluation of RAMS-results at the 19-m level (i.e., lowest RAMS half-grid wind and temperature level) against 1–10 June 2002 (first 10-days of the “present” simulation) hourly observed 2-m temperatures and 10-m wind speeds from 12 SoCAB METAR weather stations (Figure 5) was carried out by use of RAMS values at the grid point closest to each site. Vertically interpolated RAMS values yielded only small changes, and thus this correction was not calculated. Daily averaged surface temperatures from 15 COOP station were used to evaluate the past (1966–70) simulation results, as no hourly observations are available for that period. Both the RAMS 19-m level values, and all observed values, are hereafter referred to as “surface” values.

Figure 5.

NOAA 30 m resolution LULC-classification map (i.e., ocean is blue, urban brown, and non-urban green), where contours represent topography (see Figure 1 for values), plus signs represent maximum topographic heights, black squares are 12 METAR sites and blue dots are 15 COOP sites.

[30] Spatial distributions of statistical significance-levels of RAMS-produced temperature and wind- differences (i.e., present minus past) in Domain-2 were also calculated. Mean-values and standard deviations of RAMS values were thus calculated for each Domain-2 grid point at 1200, 1400, and 1600 LT over each five year past and present simulation-period for use in Student-t significance tests of differences. The study focused on these hours, as they are the period of strongest sea-breeze impacts on coastal temperatures.

3. Results

3.1. Model Evaluation

[31] Present-case RAMS temperatures (Figure 6, top ) generally compare well with observations (coefficient of determination (R2) = 0.87, Figure 7), as they capture diurnal cycles and day to day trends in peak-values; the 10 day average observed value was 19.1°C, while the modeled value was 20.3°C. RAMS captured the: large-scale cooling trend over the first three days, warming trend over the next four days, and final three-day cooling trend. Largest discrepancies occurred on the warm days, with overestimations of both their maxima by 2.5°C and minima by 2.0°C.

Figure 6.

Modeled (red) versus observed (blue) hourly averaged (over the 12 METAR stations in Figure 5 for 1–10 June 2002): (top) temperatures (°C) and (bottom) wind speeds (m s−1).

Figure 7.

Correlation of Figure 6 temperatures (gray, °C,) and wind speed (black, m s−1) data.

[32] Present-case RAMS wind speeds (Figure 6, bottom) also compare well with observations, as they again generally capture diurnal cycles and daily peak values (R2 = 0.8, Figure 7); the 10 day average observed speed was 2.9 m s−1, while the modeled value was 3.1 m s−1. Discrepancies exist, however, during the three hottest nights, which had observed near-calm winds versus an RAMS minima of about 1 m s−1; such stable SBL overestimations of near-calm speeds are common in meso-met models, and are probably related to deficiencies in their PBL schemes [Baklanov et al., 2011].

[33] Past-case JJA daily maximum surface RAMS temperatures (Figure 8) also generally compare well with observations (R2 = 0.7, Figure 9), as they capture day to day trends in peak values, i.e., large-scale cooling over the first 13 days, followed by warming over the next 13. Larger discrepancies again occur on the warm days, with overestimation of maxima by 2.5°C and minima by 2.0°C; the average observed Tmax value was 26.5°C, while the modeled was 27.3°C.

Figure 8.

Modeled (black) versus observed (gray) 1970 summer daily summer maximum temperatures (°C) averaged over the 15 COOP stations in Figure 5.

Figure 9.

Correlation plot of Figure 8 temperature data.

3.2. Climate Change Results

[34] Run-1 (present) results for Domain-1 (Figure 10, top) shows: surface mesoscale temperatures over the SoCAB basin at 1200 LT coldest over the ocean (down to 13°C), warmest inland (up to 30°C), a cool coastal strip, and cool higher-elevations east of the Basin. Up-slope winds (up to 2 m s−1) have developed over the mountains, while sea breezes (up to 2.5 m s−1) are confined to a narrow coastal strip. A near-calm region is located between the sea breeze and upslope winds, consistent with other SoCAB modeling studies for such hours [Boucouvala et al., 2003].

Figure 10.

Run-1 (present) summer-averaged 19-m Domain-1 RAMS temperatures (°C) and wind speeds (1 barb is 1 m s−1) at (top) 1200, (middle) 1400, and (bottom) 1600 LT; box represents Domain-2 area, large arrows represent average over-ocean background-flows, and dashed lines are key topographic heights.

[35] By 1400 LT (Figure 10, middle), the cool mountain tops persist, but inland temperatures have warmed (up to 32°C) and the cool coastal-strip (now at 23°C) is somewhat eroded from its value 2-h before. The sea breeze is, however, now strengthened (now up to 3.5 m s−1), penetrated further inland, and joined with the upslope winds on the eastern slopes. Two hours later (Figure 10, bottom), the cool coastal-strip and mountain-tops both still persist, while inland temperatures have reached their maximum (up to 35°C). Concurrent sea-breeze and upslope-wind penetration and intensity have also peaked (both up to 5 m s−1). Run-2 (past) results are not shown, as they look superficially similar to those of Run-1. Significant differences do exist, however, but can best be seen in the following “difference” plots.

[36] The Domain-1, 1200 LT Run-1 (present) minus Run-2 (past) difference-plots (Figure 11, top) show expected large-scale warming patterns, i.e., land temperatures have increased faster over the 35-year period than have ocean values (4.0 versus 0.5°C). The SoCAB shows the smallest warming, even cooling in a narrow coastal strip. Speed increases (as these difference-vectors are in the same general direction as the speed vectors of Figure 10) over the 35-year period in the sea breeze and upslope flows are largest over the ocean and inland mountain tops (up to 0.6 m s−1), while they are disorganized and small (only up to 0.2 m s−1) over SoCAB coastal areas, whose changes are better resolved in the Domain-2 results below.

Figure 11.

Same as Figure 10 but for differences (present minus past), where 1 barb is 0.4 m s−1 and large black arrows represent Figure 10 over-ocean background-flows.

[37] At 1400 LT (Figure 11, middle), coastal cooling over the SoCAB is better defined, while its speed differences have slightly strengthened (up to 0.4 m s−1) over the 2-h. By 1600 LT (Figure 11, bottom), speed differences have started to decrease from their peaks 2-h before, but coastal-cooling has maximized. The isolated over-ocean cooling at all three times (in the small area in the northwest of the domain) is not part of this “reverse-reaction” coastal-cooling, but part of a large-scale cooling in the NCEP BCs; discussion of its cause is beyond the scope of the current paper, but are discussed by Lebassi [2010].

[38] In summary, the coarse Domain-1 results captured many important aspects of the observed coastal cooling (e.g., strengthened sea breeze, which resulted from increased temperature gradients that resulted from a smaller large-scale warming over the ocean than over inland areas).

[39] Domain-2 difference-fields at 1200 LT (Figure 12, top), which show more details than those of Domain-1 (Figure 11) because of its finer resolution, again show that over the 35 year period, the ocean warmed less than inland areas. Peak warming-values are smaller than in Domain-1, as Domain-2 does not extend into the peak-warming inland-areas. The urban part of the coastal SoCAB, however, has only slightly warmed over the 35-year period (up to 0.5°C), as the increased sea-breeze induced coastal-cooling has partially countered its large-scale warming, even producing two small pockets of weak coastal-cooling (up to −0.5°C).

Figure 12.

Same as Figure 11 but for Domain-2, where 1 barb is 0.5 m s−1, black arrows represent Figure 10 coastal background-flows, and red lines represent urban outline.

[40] Sea breeze accelerations over the 35-year period in offshore coastal areas at this hour are more organized, and have larger vector-differences (up to 1.25 m s−1), than in the Domain-1 results. Smaller increases over the period, however, have occurred over the central urban-area than over the more rural land-area north of the city (0.75 versus 2.25 m s−1) because the large urban surface roughness (z0) minimizes the increased sea breeze flow (that formed due to large-scale warming) over the period.

[41] By 1400 LT (Figure 12, middle), the 35-year acceleration of the over-ocean flow from the Pacific High in the Domain-2 is strengthened over the last 2-h (difference vectors now up to 1.75 m s−1), as has the acceleration of the sea-breeze flow over the urban center (now up to 1.25 m s−1). Coastal cooling over the 35-years at this time at its maximum (between −0.8 and −1.0°C) than 2-h ago and now extends over the entire coastal strip.

[42] Two hours later (Figure 12, bottom), the 35-year over-ocean onshore acceleration of the previous 2-h has become an offshore directed acceleration. The coastal-plane sea-breeze acceleration is now weak and disorganized, as is the flow in areas between it and the still-organized up-slope flow. Coastal cooling has weakened, but is at its maximum inland-penetration, filling the San Fernando Valley, but stopped at the Chino and Santa Ana hills and at the San Gabriel Mountains; inland large-scale warming values over the 35-years continue to reduce from those 2-h before.

[43] Two-tailed t-tests on the 35-year RAMS Domain-2 changes in temperature (Figure 13, top) at 1400 LT (as well as at the other two hours, not shown) indicate that the coastal-cooling and (over ocean and inland) large-scale warming changes were significant at the 99% level, with less significant (<90%) values only at the boundary between the two areas. Both speed-component changes were also generally significant at 99% (Figures 13 (middle) and 13 (bottom)), except over a small coastal ocean area in the southern part of the domain, where differences are small.

Figure 13.

Domain-2 two tailed statistical significance plots at 16 LT for (top) temperature, (middle) east-west wind-component, and (bottom) north-south wind-component.

[44] The 13 western-most observed SoCAB cooling sites in Figure 1 are within the RAMS Domain-2 coastal-cooling area (including six of the seven highest-significant sites), while the four eastern most cooling sites are not. Nine of the 11 observed warming sites are likewise in the RAMS warming region (including three of four highest-significant sites). The remaining two observed warming sites are along the coast, southwest of the Santa Ana Mountains (which are parallel to the coast, just north of 117.5°W longitude), i.e., southeast of the RAMS cooling area. The RAMS coastal cooling area thus extends too far southward along the coast, but not far enough eastward in the inland area between the Santa Ana and San Bernardino Mountains (at the northeastern corner of Domain 2). The first effect resulted because of the 4 km Domain 2 grid spacing could not resolve the true elevation of the Santa Ana Mountains. The modeled coastal cooling flow was thus not able to be blocked by their true elevation, and thus the modeled flow went north (and not south) of them.

[45] In summary, Domain-2 fine-scale surface 35-year summer daytime changes better resolved (as compared to the coarser Domain-1 results) many important aspects of the observed coastal-cooling. These include its location, magnitude, and strengthened sea breezes, which resulted from increased temperature-gradients from the lower over-water large-scale warming rates, as compared to those over inland areas.

4. Conclusion

[46] The meso-met RAMS model was used to investigate local (on 16 and 4 km grids) climate changes in the SoCAB, with simulations designed to quantify impacts from large-scale warming. Results from a simulation with present (2001–05) summer climate-conditions and present LCLU-patterns were compared to one with present-LCLU and past (1966–1970) climate-conditions.

[47] Evaluation of the RAMS results against hourly SOCAB surface-temperature and -wind observations during a current 10-day “present” summer period over 12 METAR stations showed that they generally captured both diurnal cycles and day to day trends in peak and minimum values; domain-averaged mean-biases were only 1.2°C and 0.2 m s−1, respectively. Evaluation against “past” daily surface maximum temperatures averaged over 15 COOP stations also showed good agreement in trend and magnitude, with a domain-average bias of only 0.7°C.

[48] Present-case RAMS results for the coarser outer domain showed noon summer SoCAB surface temperatures coldest over the ocean and warmest inland, with a cool coastal strip and mountain peaks. They also showed upslope winds over the mountains, sea breezes confined to a narrow coastal strip, and a near-calm region between the two. During the next four hours, the cool mountain tops persisted; inland temperatures increased; and the sea breeze strengthened, penetrated further inland, and joined with the upslope winds on the eastern slopes.

[49] RAMS coarse-domain (present minus past) differences showed the expected large-scale warming over the 35-years, i.e., land temperatures increased faster than coastal ocean SST values (up to 2.0 and 0.5°C, respectively). The SoCAB showed the least warming, however, with coastal cooling in a narrow strip, as cooling from the increased sea breeze was greater than the large-scale warming. Wind speed increases in both the sea breeze and upslope flows were largest over the ocean and mountain tops (both up to 1.6 m s−1) and smallest over the SoCAB (only up to 0.2 m s−1). Two hours later, coastal cooling was maximized (up to −1.6°C), while speed differences had started to decrease.

[50] Fine-domain (present minus past) differences better resolved the maximum magnitude (still at 1400 LT) 35-year SoCAB coastal cooling (greater than −0.8°C, but less than −1.0°C) and sea breeze acceleration (up to 1.5 m s−1). Two hours later, coastal-plain sea breezes became disorganized, and while coastal-cooling magnitudes decreased, this hour showed its maximum inland penetration (over the entire coastal strip). Maximum simulated SoCAB coastal cooling over the last 35-years was thus at the location found in the observational phase of the study and at almost its magnitude (−0.99°C decade−1). The RAMS maximum cooling is less than the observed value because it is a difference and not a trend and because its 4 km grid resolution cannot resolve the tight gradient in the cooling values around the single station with this observed large value.

[51] In summary, the coarse-domain RAMS results captured many important aspects of the observed surface inland-warming and concurrent coastal-cooling that developed during the 35-years between the present and past summer-daytime simulation periods, i.e., sequentially: weaker large-scale warming over the ocean than over inland areas, increased surface temperature gradients, strengthened sea breezes, and thus coastal cooling. The fine domain results further resolved the magnitude and spatial extent of these impacts.

[52] The coastal-cooling reverse-reaction to large-scale warming found in the current study could thus be expected in all subtropical low-elevation west-coast Koeppen Marine-Mediterranean (i.e., cool dry-summer subtropical) Csb climate areas (i.e., in coastal California, Chile, Peru, Australia, South Africa, and Portugal), in which westerly directed sea breezes strongly influence regional climate. The modeling study by Comarazamy et al. [2010] thus showed that in the Caribbean region east-coast city of San Juan, Puerto Rico, where coastal areas are dominated by easterly flows, such cooling did not occur.

[53] Intergovernmental Panel on Climate Change [2001] annual temperature trends for 1976–2001 do, in fact, show such cooling (up to 0.6°C decade−1) at all the west coast areas listed above, except Portugal. Recent local observational studies have also detected coastal cooling; e.g., Falvey and Garreaud. [2009] analyzed 1979–2006 NCDC and gridded observed (annual-average and maximum) temperature trends over coastal Chile, and found coastal-cooling. Coastal sediment-cores along Peru, analyzed by Gutiérrez et al. [2011], also showed cooling SSTs for the latter part of the 20th century.

[54] Oglesby et al. [2010] analyzed 4 km WRF modeled-differences (between the periods 2000–2004 and 2050–2054), forced with CCSM output, for Meso-America; coastal cooling was predicted along the Pacific coasts of Mexico and Central America. This result indicates that the currently discovered SoCAB coastal cooling might likewise continue over the next 50 years, although a similar series of WRF simulations by Zhao et al. [2011] for all of California did not show coastal-cooling, perhaps because they forced their simulations with output from the older Parallel Climate Model (PCM) model.

[55] Significant beneficial societal impacts from this reverse-reaction to large-scale warming include possible decreased maximum: O3 levels, per-capita energy requirements for cooling, and human thermal-stress levels. Future modeling efforts could enhance the current simulations by inclusion of: longer continuous-simulation periods; finer scale horizontal grid resolutions; updated CO2 concentrations; advanced urbanized parameterizations, such as that of Martilli [2002]; larger horizontal domains; El Nino, aerosol radiative feedbacks; upwelling and coastal fog; updated NCEP data sets; detailed time and space varying SSTs and upwelling changes; station by station statistical evaluations, future-impact evaluations; and the full diurnal and annual cycles of these interactions.


[56] The authors would like to thank Edwin Maurer and Drazen Fabris of Santa Clara University (SCU) and Cristina Milesi of the California State University Foundation, Monterey Bay, at NASA Ames Research Center, for their insightful comments. We also thank the School of Engineering, at SCU and the National Science Foundation grant 0933414 for funding the lead author. We also acknowledge the City College of New York and San Jose State University for providing the computational time.